Abstract—An algorithm for tracking multiple objects using locally adaptive generalized filtering is proposed. The tracking algorithm is invariant to geometric transformations of objects, partial occlusion of objects, nonuniform illumination of scene, and additive noise in scene images. The proposed system utilizes generalized optimal correlation filters and a prediction scheme based on the kinematic model of objects motion. With the help of iterative training, the training system can be adapted to current scene changes. The performance of the proposed algorithm is compared with that of the state-of-the-art visual tracking algorithms on known databases in terms of commonly accepted quality metrics and processing time.

摘要—提出了一种使用局部自适应广义滤波来跟踪多个对象的算法。跟踪算法对于对象的几何变换，对象的部分遮挡，场景的不均匀照明以及场景图像中的附加噪声是不变的。提出的系统利用广义最优相关滤波器和基于物体运动学运动模型的预测方案。借助迭代训练，可以使训练系统适应当前的场景变化。就普遍接受的质量指标和处理时间而言，将所提出算法的性能与已知数据库上的最新视觉跟踪算法的性能进行比较。